Ames Laboratory, Department of Energy, ISU Ames, Iowa
Fastest Real Application award

David Deaven, Kai-Ming Ho, Dave Turner, Jamie Morris
SuperComputing 1995

          David Deaven and Kai-Ming Ho have developed a new Genetic Search Algorithm designed to find the atomic arrangement that has the lowest total energy. The genetic search starts with randomly generated configurations of atoms as 'parents' and then combines various parts of different 'parents' to form 'offspring'. Each of the offspring is then relaxed using our Classical or Tight-Binding codes.Then, the lowest energy offspring are used as parents for the next generation. By repeating this cycle for thousands of generations, we efficiently search avery wide variety of configurations.

This algorithm is ideal for running on massively parallel computers or even on a distributed set of computers.

          This work recently won the SuperComputing '95 Fastest Real Application award as part of the High Performance Computing Challenge. We used the I-WAY to connect a variety of supercomputers spread across the country in order to perform a single run of the Genetic Search Algorithm.We achieved a sustained computational rate of over 100 GFlops, and we could easily exceed this if given access to more supercomputers.

For more information, contact
David Deaven, Kai-Ming Ho,
Dave Turner, or James Morris. The URL for this document is http://www.scl.ameslab.gov
Revised